Gene network biological validity based on gene-gene interaction relevance

Abstract

In recent years, gene networks have become one of the most useful tools for modeling biological processes. Many inference gene network algorithms have been developed as techniques for extracting knowledge from gene expression data. Ensuring the reliability of the inferred gene relationships is a crucial task in any study in order to prove that the algorithms used are precise. Usually, this validation process can be carried out using prior biological knowledge. The metabolic pathways stored in KEGG are one of the most widely used knowledgeable sources for analyzing relationships between genes. This paper introduces a new methodology, GeneNetVal, to assess the biological validity of gene networks based on the relevance of the gene-gene interactions stored in KEGG metabolic pathways. Hence, a complete KEGG pathway conversion into a gene association network and a new matching distance based on gene-gene interaction relevance are proposed. The performance of GeneNetVal was established with three different experiments. Firstly, our proposal is tested in a comparative ROC analysis. Secondly, a randomness study is presented to show the behavior of GeneNetVal when the noise is increased in the input network. Finally, the ability of GeneNetVal to detect biological functionality of the network is shown.

Figure 1

A schematic representation of GeneNetVal methodology. In the first step, organism o's information is extracted from KEGG database. Each of the M metabolic pathways is processed to obtain M gene networks. In the second step, M evaluations of the input network are carried out. Note that the results presented were obtained by applying our approach at level 1.

Figure 2

The simplest conversion example. In the first substep the compound nodes and the direction of the relationship edges are removed. In the second substep, new association relationships are established.

Figure 3

Two portions of real KEGG pathways with multiconnected nodes. (a) contains a fragment of the pathway hsa00410 where three genes are connected to the same compound. In the process of conversion to a gene network, new relationships between these genes are created. (b) shows a fragment of the sce00510 pathway, illustrating how a compound is transferred by two genes. When the gene network is created, these two genes must be connected (as shown in the figure). Note genes HSD11B1, AKR1D1, and AKR1C4 (a) correspond to enzymes “1.1.1.146,” “1.2.1.3,” and “1.1.1.50,” respectively.

Figure 4

An example of the comparison using level 1 and level 2. Examples of Hit₁ and Hit₂ are presented. The purple nodes and their relationships are pruned for this specific evaluation because they do not belong to the metabolic pathway.

Figure 5

Representation of a toy example for the ROC study performed. (a) represents the GeneNetVal process, where the validity values for both networks are obtained. In (b) the confusion matrices are obtained. The TPR and FPR values are presented in (c). Finally the ROC curve is depicted in (d).

Figure 6

ROC analysis of our methodology using some yeast networks. For this analysis two different topologies were used: pure random and scale-free topology.

Figure 7

Results of the randomness study of GeneNetVal using level 2. For this study, we have used different yeast networks versus pathway sce04111.

Algorithm 1

Pruning process.